There is a longstanding recognition that asthma is a variable disease and attempts to elucidate the causes of asthma have been hampered by its genetic and phenotypic heterogeneity. The recognition that asthma constitutes several distinct phenotypes, the development of novel non-invasive biomarkers of airway inflammation and the desire to maximise information from large-scale genotyping studies have prompted new approaches to defining asthma phenotypes in adults and children. These have included statistical modelling of longitudinal symptom data in epidemiological studies and a re-examination of combinations of clinical, physiological and pathological markers that signify discrete disease entities. It is hoped that better understanding of asthma phenotypes will provide useful new insights into asthma aetiology but will also be of immense benefit in developing and tailoring asthma therapy to individual patients. This review aims to summarise traditional approaches to categorising asthma into different types and to consider novel and emerging approaches to this problem and their likely impact on understanding the causes and natural history of asthma.

PHENOTYPE: WHAT’S IN A NAME?

A phenotype refers to a set of characteristics that can be used to classify organisms into discrete groups. The Danish botanist and geneticist, Wilhelm Johannsen coined the terms “genotype” and “phenotype” in an address to the American Society of Naturalists in 1910.1 Phenotype was interpreted to designate “a group of organisms, which in outward appearance seemed to belong to one type”.2 This was clarified in correspondence from Shull, who noted that the term phenotype referred to “the constitution or assemblage of characteristics” with respect to which a group of individual organisms is apparently homogeneous and not to the group of individuals themselves.3 Therefore, phenotype can refer to any observable characteristic of an organism, including morphology, development, biochemical or physiological properties, and behaviour. Each phenotypic characteristic will depend to a greater or lesser degree on genetic heredity and environmental influences. The search for modifiable factors that increase the risk for asthma is moving towards studies of gene–environment interactions.4 Careful phenotype classification is important to the detection of causal pathways that may apply only to subgroups of asthmatics and in translating the results of such studies to clinically relevant populations.

DEFINING ASTHMA AND RELATED PHENOTYPES: TRADITIONAL APPROACHES

Paediatricians will generally feel comfortable with the diagnosis of asthma in a clinical setting, in which clear guidelines exist.5 In spite of this, it is increasingly accepted that the diagnosis of asthma is a probability-based assessment that may or may not be supported by corroborative evidence from respiratory function tests or adjunct investigations. Therefore, the diagnostic approach has evolved from one incorporating the careful exclusion of other causes of airway obstruction6 7 to the recommendation in the current British Thoracic Society/Scottish Intercollegiate Guidelines Network guidelines that the variable likelihood of asthma in children is acknowledged in the diagnosis by assigning high, low or intermediate probability of the disease.5 Information from additional investigations is seldom readily available in young children but even this can complicate questions of disease classification. Airway inflammation, atopy, and bronchial hyper-responsiveness are all recognised constituents of asthma but the question of whether these features are independent pathophysiological entities or part of a spectrum of the same disease is not fully resolved. Terms, such as “reactive airways disease” have emerged that almost certainly embrace components of asthma, in this case bronchial hyper-reactivity, but which are not synonymous with the paradigm recognised in clinical practice. The incorporation of this concept into “reactive airways dysfunction syndrome” is now almost exclusively used in the context of occupational or toxic exposures.8

Although much effort has been expended in refining diagnostic criteria of asthma to improve diagnostic accuracy, there has been a longstanding recognition that asthma is not a single disease entity but incorporates a number of clinical syndromes. This has led to repeated calls for the abolition of the term asthma altogether, as it has been seen as an impediment to progress in understanding the natural history and causes of different subtypes of asthma.9 10 Several approaches have been used to separate these phenotypes, based around symptoms and triggers of wheezing illness or pathophysiological features of the disease.

SYMPTOMS AND TRIGGER-BASED CLASSIFICATION

Early epidemiological studies of asthma or wheezing illnesses in childhood recognised that different patterns of wheezing (in terms of age of onset, frequency, severity and triggers) had differing natural histories.11 One of the best recognised examples of longitudinal studies of asthma started in Melbourne in 1967 and challenged the prevailing paradigm of differentiation of early childhood wheeze into asthma and wheezy bronchitis, arguing that these were part of the same spectrum of illness.12 However, wheezy bronchitis or viral-triggered wheeze has been extensively studied, notably by Silverman and colleagues in Leicester,13 culminating in the recent publication of a consensus approach to the diagnosis and management of wheezing in young children that conceptualises early life wheezing as episodic (viral) wheeze and multi-trigger wheeze.14 Follow-up studies of infants with wheezy bronchitis have suggested that this entity has distinct characteristics, including physiological and clinical outcomes that are more closely related to chronic obstructive pulmonary disease than to asthma.15

A concept analogous to wheezy bronchitis also emerged from the Tucson Children’s Respiratory Study, which categorised wheezing illnesses in early childhood into discrete phenotypes based on longitudinal surveys of wheezing patterns.16 The history of reported wheezing was used initially to define three phenotypes of wheeze that evolved during the first 6 years of life: transient early wheeze (starting before 3 years and remitting by age 6 years), late-onset wheeze (not present by 3 years but reported at 6 years) and persistent wheeze (wheeze in the first 3 years that persisted to age 6 years). Continued follow-up of the Tucson cohort has enabled the investigators to refine their ideas about these early wheezing patterns and to conceptualise them as: transient wheeze in infancy associated with reduced lung function soon after birth and characterised by wheezing episodes in the context of viral respiratory infections with resolution in the pre-school years; non-atopic wheeze starting in infancy, persisting into mid-childhood and associated with absence of allergic sensitisation; and asthma, which increases in prevalence after infancy, is associated with normal lung function soon after birth, a positive family history and personal evidence of atopy.17 These phenotypes have served as a useful model for over a decade and the success of the investigators in following the cohort through adolescence has led to a number of important observations and discoveries in both conventional, observational epidemiological studies and the genetic epidemiology of asthma and wheezing in childhood (reviewed by Taussig18).

However, the Tucson paradigm gives an incomplete description of the totality of wheezing illness in early childhood and there are data that are not consistent with the concepts proposed. For example, heterogeneity of T-cell immunity, even within a seemingly homogeneous group of atopic asthmatics has been reported,19 echoing Johannsen’s observations in 1911 that “there may be found considerable genotypical differences hidden in apparently homogenous populations”.1 Our group reported divergence of wheezing phenotypes at an earlier age than reported in the Tucson study20 and more recently lung function abnormalities in infancy have been found to predate asthma in later childhood.21

PATHOLOGY OF AIRWAY INFLAMMATION

Airway inflammation is a major component of asthma pathology but difficulties in accessing tissue samples from the lungs have been a barrier to understanding the relationships between clinical and pathological features of asthma in young children. Endobronchial biopsy with histopathological classification of airways inflammation has been widely utilised in the study of asthma in adults for many years.22 This technique, coupled with bronchoalveolar lavage, has provided insights into airway inflammatory processes and changes in the basement membrane and smooth muscle cells suggesting airway remodelling in the context of airways inflammation.23 These techniques have also been applied to studies of children, in whom they have been reported to be safe and informative,24 revealing similar evidence of inflammation and structural airway changes, even in preschool wheeze.25

Histopathological and ex vivo cellular studies have also been applied to the study of “difficult-to-control asthma” in adults and children. This phenotype has been the subject of increasing attention, both from an aetiological and therapeutic perspective.26 It is a complex entity with many contributing factors beyond the scope of this review, including adherence to therapy, psychological variables, comorbidities and environmental influences. However, there is strong evidence that a subgroup has severe asthma and this group may include individuals with corticosteroid insensitivity at cellular level.27 There is evidence from adult studies that at least two different clinical and pathological subtypes of difficult-to-control asthma exist28 and there is evidence of a similar dichotomy in children.29 Histopathological studies of bronchial biopsies from children with difficult asthma30^–32 show relationships of inflammatory cell types, basement membrane and airway smooth muscle changes with clinical and physiological outcomes. Another explanation for differential responses to treatment could lie in genotypic variation in pharmacological receptors, such as the beta-adrenergic receptor. Polymorphisms of this gene have been associated with a higher risk for asthma33 and also with responses to beta agonists in acute asthma34 and with increased exacerbations in children treated with long-acting beta agonists.35

NEW APPROACHES TO DEFINING ASTHMA PHENOTYPES

As noted above, a full understanding of the inflammatory events associated with asthma in children has been hampered by limitations in obtaining tissue samples from the airways. However, non-invasive markers of airway inflammation that are applicable to large-scale studies of wheezing children have been developed. The principal methods used are analysis of induced sputum, measurement of fractional exhaled nitric oxide (F_ENO) and measurement of a variety of analytes in exhaled breath condensate. These have proven utility in monitoring inflammation and response to therapy in asthma,36 37 as well as having the potential to aid further dissection of the asthma phenotype by at least proxy measures of the predominant inflammatory pattern. This science is at an early stage in children, although the greater availability of data in adults demonstrates how a variety of clinical, physiological and inflammatory markers can be combined to create a sophisticated paradigm of airway inflammatory diseases that encompass the major clinical syndromes from asthma through to chronic obstructive pulmonary disease.38

Moeller and colleagues have recently shown that exhaled NO (F_ENO), regarded as a marker of eosinophilic airway inflammation, is well correlated with predictive markers of asthma in preschool children with wheeze.39 Another study of F_ENO in infants reported raised levels preceding the onset of respiratory symptoms in the offspring of atopic mothers, suggesting that the presence of airway inflammation is an early event in the evolution of asthma and wheezing illnesses.40 This accords with other data which support associations between lung function21 and bronchial responsiveness41 soon after birth and subsequent development of asthma in later childhood.

Further insight into airway inflammation is likely to arise from studies of biomarkers in moisture condensed from exhaled breath using a cooling system. The condensate contains a number of inflammatory markers (reviewed by Liu and Thomas42), some of which have been associated with asthma in children.43^–45 At present they appear to be of more use in distinguishing asthma from non-asthma than in helping to characterise discrete phenotypes. However, with greater experience of these methods and their application to large, population-based studies46 and the development of new biochemical methodologies such as metabolomics,47 this situation is likely to evolve rapidly.

A different approach to phenotype definition has been based around applying novel statistical modelling methods to collections of asthma features, including symptoms and objective measures of asthma-related phenotypes, such as atopy and lung function measures. Latent class analysis (LCA) is a form of cluster analysis in which subjects are grouped according to similarities of response levels to a number of measured categorical variables. The smallest number of groups (latent classes) that describe the variance of responses to these variables within the population is estimated by the model. This tool has been widely applied to analysis of market research, sociological, and psychological data and is increasingly applied to health-related data to discover case subtypes and to evaluate diagnostic tests. We used LCA in the Avon Longitudinal Study of Parents and Children (ALSPAC), a large birth cohort in the UK, to derive subtypes of wheezing based on longitudinal trajectories of repeat assessments of wheeze at annual intervals. This approach yielded five discrete wheezing phenotypes typified by age of onset and persistence of wheezing symptoms over the first 6 years after birth. Three of these were concordant with the Tucson data; transient early wheeze, late-onset wheeze and persistent wheeze but two novel phenotypes emerged (together comprising 12% of the total population); intermediate-onset wheeze with onset around 18 months and prolonged early wheeze that was present at 6 months and persisted beyond 5 years. These differed in their associations with lung function, bronchial responsiveness and atopy measured at 7–9 years, hinting that they may represent discrete pathophysiological entities with differing aetiologies and natural histories.48 In the case of the analyses above, the multiple variables were wheezing at different time points. Spycher and colleagues used LCA in the Leicester birth cohorts to derive phenotypes based on symptoms, lung function, airway responsiveness and skin prick tests in preschool children.49 Smith and colleagues have argued that focusing on a single symptom, such as wheezing or doctor-diagnosed asthma used in many epidemiological studies, fails to capture the complexity of the asthma phenotype and its relationship with underlying pathophysiological processes. They used principal components analysis of multiple parent-reported respiratory symptoms at ages 3 and 5 years in the Manchester Asthma and Allergy Study to derive syndromes of co-existing symptoms and tested their associations with objective outcomes.50 Principal components analysis is a method that transforms a number of variables that may be related to each other into a smaller series of uncorrelated variables (principal components), such that the first principal component explains as much of the variance in the data as possible, the second component as much of the remaining variance as possible and so on. These authors described a four-component solution to 21 questionnaire items at age 3 years comprising wheeze, cough, colds and chronic symptoms, and a five-component solution to 32 questionnaire items at 5 years comprising wheeze, cough, wheeze with irritants, wheeze with allergens and chest congestion. The patterns of associations of these components with objective measures such as airway reactivity and with markers of atopic status were interpreted as suggesting different pathophysiological processes contributing to each component. For example, airway resistance at age 3 years predicted both the wheeze and cough components but not colds or chronic symptoms.

The above approaches are attractive as they begin with little preconceived notion of phenotype number or structure but have produced clinically meaningful groups related by common phenotypic characteristics and distinct from other phenotypes in their relationships with objective and other markers of asthma. However, establishing the link between characteristics of a latent class or principal component and an observable pathological phenotype, which could be regarded as a gold standard for identification of different inflammatory phenotypes of asthma, is some way off. Also, reliant as they are on the availability of existing data, these techniques should not yet be regarded as having utility in assigning phenotype membership to individual patients for the purpose of predicting natural history of asthma or determining effective therapeutic interventions.

CLINICAL AND RESEARCH IMPLICATIONS

There is a continued need for the identification of clinically relevant phenotypes of asthma, particularly during childhood when many of the major influences on asthma development appear to operate.51 Genome-wide analyses have been successful in identifying new genes associated with asthma.52 However, a recent meta-analysis of genome-wide linkage studies for asthma and related traits found evidence for linkage of chromosomal regions to intermediate, quantitative phenotypes, such as bronchial responsiveness but not to asthma.53 If we are to maximise the potential of genetics to answer important questions about asthma aetiology, it is important to consider the impact of precise classification of phenotypes on issues such as replication of associations, misclassification bias and harmonisation of outcomes across populations large enough to achieve sufficient statistical power, particularly for the detection of gene–gene and gene–environment interactions.54

The approaches described in this paper have gone some way towards this but questions remain about the relationship of clinically defined phenotypes based on symptoms, triggers and their natural history with pathological processes, such as eosinophilic and non-eosinophilic airway inflammation, and physiological measures, including airway function and responsiveness to bronchial challenge (summarised in table 1). It is likely that there will be considerable overlap in these relationships between different clinical syndromes, as reviewed by Wenzel.55 An example where childhood and adult phenotypes might overlap is evidenced by the few opportunistic studies of viral-associated wheeze in young children that have demonstrated predominantly neutrophil- as opposed to eosinophil-mediated airway inflammation.56 57 These results suggest the presence of different pathological phenotypes of wheezing illness in young children and also hint at similarities with adult classifications of non-eosinophilic airway inflammation.38

Improved classification of asthma phenotypes in children will be expected to increase our understanding of the aetiology and natural history of wheezing illnesses in childhood and beyond. This may indicate novel strategies for primary and secondary prevention of disease in specific subgroups and is also important in the development of pharmacological treatments of asthma, where selected treatments may be unfavourable to some phenotypes35 and phenotype-specific treatment strategies will need to be adopted, as suggested for difficult-to-control asthma.58